2.2.2 (2026-03-28)
- add interleaving check to deduplication to preserve conversational turn ordering (5ef8b2f)
- bump pyjwt and orjson to fix HIGH-severity CVEs (9489e85)
- extract shared tool infrastructure and resolve batch-nonbatch parity bugs (f337a67)
- pass API keys explicitly in summary providers and clean up project config (ff03172)
2.2.1 (2026-03-26)
- resolve hybrid search quoted phrase loss, embedding guard asymmetry, and batch content_type drift (069599b)
- serialize SQLite write queue operations through writer lock (10db4cc)
2.2.0 (2026-03-25)
- add site-packages path trimming to universal logger (b5088ff)
- update default OpenAI summary model to GPT-5.4 Nano (1dfa0cd)
- prevent reasoning models from exhausting summary output token budget (347cfc0)
2.1.0 (2026-03-18)
- add automatic model pulling for Ollama providers on startup (3196c2b)
- add context window validation for summary generation (7ea909e)
- add eager model loading and concurrency protection for FlashRank reranking (f260d27)
- add Ollama cold-start optimization with model pre-warming (41e2776)
- add SHA-256 content hash for deduplication optimization (76ec45f)
- add Skill Integration section to DEFAULT_INSTRUCTIONS (7b84ae0)
- change default summary model to qwen3:0.6b (c5c7c0e)
- classify PostgreSQL initialization errors and limit Docker restart retries (ef6e58a)
- eliminate redundant PostgreSQL schema execution (976d625)
- enforce uniform Generation-First Transactional Integrity across all tools (a1bb652)
- replace per-call-site JSON string deserialization with schema-aware FastMCP middleware (f706b5b)
- replace provider-availability checks with actual generation counters in batch tool messages (a473919)
- resolve <unknown> in tenacity retry log messages (91e4bb4)
- resolve Docker and Helm deployment issues with model pull and summary support (1c893d1)
- use source-aware dynamic prompts for summary generation (5c80bea)
2.0.0 (2026-03-13)
- This release introduces two breaking changes:
- add LLM-powered summary generation for context entries (9f4c9e2)
1.7.1 (2026-03-09)
- change FASTMCP_STATELESS_HTTP default from false to true (277baa3)
- prevent event loop blocking in FlashRank model loading (b0076fd)
1.7.0 (2026-03-08)
- add configurable micro-batching for FlashRank reranking (43fcd9c)
- apply ts_headline only to LIMIT'd rows in PostgreSQL FTS (2cdb52e)
- improve tool description clarity and self-sufficiency (ed522c3)
- offload FlashRank ONNX inference to thread pool via asyncio.to_thread (272eb9e)
- report project version in MCP protocol handshake (0372f58)
- skip auth initialization on stdio transport (7fbc9ef)
1.6.0 (2026-03-04)
- add adaptive FTS mode to hybrid search for improved long-query recall (54c5f31)
- add RERANKING_CPU_MEM_ARENA env var to control ONNX memory arena (5f98cf7)
- add temporary monkey-patches for MCP SDK session crash on client disconnect (d9a9f3d)
- correct deduplication data integrity in store_context and store_context_batch (a395ede)
- correct modes_used to reflect execution rather than results in hybrid search (c747edf)
1.5.0 (2026-03-02)
- add configurable ONNX intra-op thread control for FlashRank reranking (1352988)
- add embedding concurrency control, dynamic timeout, and search quality improvements (757d19f)
- sanitize hybrid search warning messages to avoid leaking error details (d71b0c9)
- suppress onnxruntime type errors via mypy and pyright config (e59e9de)
1.4.0 (2026-02-27)
- add FASTMCP_STATELESS_HTTP setting for horizontal scaling (c94a423)
- add MCP server instructions support (e1805a6)
- remove search_modes parameter from hybrid_search_context (2335d0c)
- remove install hints from inner exceptions in providers and services (e9708b2)
- replace session-scoped advisory locks with transaction-scoped locks in migrations (4508eb8)
- update dependencies to resolve Trivy CVE findings (c21d736)
1.3.3 (2026-02-11)
- add advisory locks to PostgreSQL migrations and schema initialization (d637cc5)
- add PostgreSQL connection resilience defensive hardening (e0ae6da)
- add ROLLBACK to PostgreSQL backend (bb69f45)
1.3.2 (2026-02-06)
- add python-multipart constraint for CVE-2026-24486 (9c6d2ea)
- resolve embedding deduplication race condition (89f6596)
- resolve PostgreSQL migration timeout and idempotency issues (d2f8f8e)
1.3.1 (2026-01-25)
- classify PostgreSQL backend errors to control container restart behavior (f5ed4c8)
- upgrade FastMCP to 2.14.4 (8780345)
1.3.0 (2026-01-20)
- add Pgpool-II detection for PostgreSQL backend (e64f082)
1.2.1 (2026-01-18)
- add backend-specific FTS tool descriptions (09113cd)
1.2.0 (2026-01-17)
- add configurable asyncpg prepared statement cache settings (5ebf287)
1.1.0 (2026-01-17)
- implement embedding-first transactional integrity (c9e4c12)
1.0.0 (2026-01-16)
- API response structure changed for all search tools. Users must update their code:
- FTS: result['score'] -> result['scores']['fts_score']
- Semantic: result['distance'] -> result['scores']['semantic_distance']
- All tools: result['rerank_score'] -> result['scores']['rerank_score']
- Default embedding model changed from embeddinggemma:latest (768 dim) to qwen3-embedding:0.6b (1024 dim). Existing vector databases will have incompatible embeddings.
- ENABLE_EMBEDDING_GENERATION now defaults to true. Server will NOT start if embedding dependencies are not available when ENABLE_EMBEDDING_GENERATION=true (the default).
- add chunk-aware reranking for FTS, semantic search, and hybrid search (ff79859)
- add embedding truncation control with universal validator (d941b6f)
- add text chunking and cross-encoder reranking (942bb24)
- implement universal retry wrapper for embedding providers (4031061)
- prevent Docker infinite restart loops with exit code handling (91acfa4)
- replace embeddinggemma with qwen3-embedding:0.6b (02ce75e)
- separate embedding generation from semantic search (3967262)
- standardize scores API across all search tools (0a940c9)
- switch PostgreSQL FTS ranking from ts_rank to ts_rank_cd (a58fd9c)
- enable LangSmith tracing for embedding operations (ba7bf9f)
0.17.0 (2026-01-11)
- implement LangChain embeddings multi-provider architecture (7382629)
- add schema qualification to recursive jsonb_merge_patch call (51bc48c)
- resolve critical PostgreSQL backend issues (8186f90)
0.16.1 (2026-01-09)
- add POSTGRESQL_SCHEMA setting and refactor metadata index management (772f3aa)
- improve error handling and add timeout/retry logic (319e08f)
- replace hardcoded public schema with POSTGRESQL_SCHEMA setting (3e6511f)
0.16.0 (2026-01-06)
- add configurable metadata field indexing with sync modes (10ab4cb)
0.15.1 (2026-01-05)
- add array_contains to MCP tool metadata_filters descriptions (45f2357)
0.15.0 (2026-01-05)
- add array_contains operator with graceful non-array handling (b23642a)
0.14.1 (2025-12-30)
- add search_path to functions for CVE-2018-1058 mitigation (d156235)
0.14.0 (2025-12-28)
- add HTTP authentication support with bearer token and OAuth options (e625b42)
0.13.0 (2025-12-27)
- add DISABLED_TOOLS environment variable and MCP tool annotations (f6bccd7)
0.12.0 (2025-12-26)
- add compose configuration for external PostgreSQL (61b294f)
- add Docker deployment with HTTP transport support (77390e9)
- split compose files for independent SQLite and PostgreSQL deployments (7840291)
0.11.0 (2025-12-22)
- standardize search tools API response structure (04d7472)
- add explain_query support to semantic search for consistent stats structure (cd5884b)
- remove duplicate Args sections from MCP tool docstrings (9e0df55)
- replace blocking pathlib.Path with anyio.Path in async code (0101b0a)
- standardize MCP tool docstrings to use inline JSON structures (7bef90d)
0.10.0 (2025-12-21)
- add explain_query parameter to fts_search_context (7334e9f)
- add explain_query parameter to hybrid_search_context (b32129d)
- add hybrid search with RRF fusion (8c6ea15)
- add offset, content_type, include_images to search tools (bb22012)
- add tags parameter to semantic, FTS, and hybrid search (bcf1332)
- add uniform backend display to all statistics output (0d7c988)
- rename top_k to limit in semantic_search_context (2c27c14)
- standardize limit parameter across all search tools (d85d07d)
- handle hyphenated words in FTS queries (b2f9ccb)
- hybrid search pagination and test quality improvements (2da98fe)
- make startup checks backend-specific (3ee1149)
- register hybrid_search_context tool dynamically (0ee0d22)
0.9.0 (2025-12-03)
- add full-text search with FTS5 and PostgreSQL tsvector (34ef13f)
- handle boolean metadata filtering for PostgreSQL JSONB (1ff407e)
0.8.0 (2025-11-30)
- add bulk operations for batch context management (7516a93)
- add metadata filtering to semantic_search_context (50346d6)
- handle integer arrays in in/not_in metadata operators (23076df)
0.7.0 (2025-11-29)
- support date filtering in search_context and semantic_search_context (e433f5e)
0.6.0 (2025-11-29)
- add metadata_patch parameter for partial metadata updates (42f7a5f)
0.5.1 (2025-11-26)
- resolve Supabase and PostgreSQL issues (3d5c450)
0.5.0 (2025-11-23)
- add PostgreSQL backend with pgvector semantic search (03ad9b0)
- add storage backend abstraction for multi-database support (4e7744a)
- eliminate duplicate tool registration and improve naming (d46cb7c)
- eliminate redundant backend initializations during startup (dcd14d7)
- load sqlite-vec extension before semantic search migration (3069460)
- resolve asyncio primitives event loop binding issue (705fd9d)
- resolve integration test hang with persistent backend (35219db)
0.4.1 (2025-10-10)
- allow nested JSON structures in metadata (7f624ee)
0.4.0 (2025-10-06)
- add semantic search with EmbeddingGemma and sqlite-vec (2e0d3db)
- enable configurable embedding dimensions for Ollama models (2d68963)
- correct semantic search filtering with CTE-based pre-filtering (66161a3)
- resolve type checking errors for optional dependencies (be47f9d)
0.3.0 (2025-10-04)
- add update_context tool for modifying existing context entries (08aed11)
- enforce Pydantic validation and resolve test reliability issues (6137efc)
- ensure consistent validation patterns across all MCP tools (7137aca)
- move validation to Pydantic models (1e2e480)
- resolve all pre-commit issues and test failures (0a2142d)
0.2.0 (2025-09-28)
- add comprehensive metadata filtering to search_context (e22cfe0)
- improve metadata filtering error handling and query plan serialization (faa25b6)
- remove REGEX operator and fix case sensitivity for string operators (b6d3534)
- add initial version
- add initial version (ac17f19)